Jake Lee
@jakehlee.com
Data Scientist @ NASA/Caltech Jet Propulsion Laboratory, Machine Learning for Earth and Space Science. EMIT Imaging Spectroscopy (Clouds & CH4), Euclid/Roman Cosmology. Volunteer robotics head coach FRC 2404.
Following up on this, here's an even cooler animation of the SWOT overpass!
earthobservatory.nasa.gov/images/15466...
earthobservatory.nasa.gov/images/15466...
August 18, 2025 at 9:59 PM
Following up on this, here's an even cooler animation of the SWOT overpass!
earthobservatory.nasa.gov/images/15466...
earthobservatory.nasa.gov/images/15466...
CGP Grey has taken "thumbnail optimization" very literally by applying common deep learning augmentation techniques to his older videos' thumbnails: color inversion, tile flipping, gaussian noise... It's certainly working, I myself absentmindedly clicked on one.
April 22, 2025 at 5:19 AM
CGP Grey has taken "thumbnail optimization" very literally by applying common deep learning augmentation techniques to his older videos' thumbnails: color inversion, tile flipping, gaussian noise... It's certainly working, I myself absentmindedly clicked on one.
New preprint! We use generative diffusion models to quickly produce realistic #Euclid -like galaxies, a step towards addressing challenges in calibrating and validating weak lensing analyses. 🔭🧪
arxiv.org/abs/2504.07183
arxiv.org/abs/2504.07183
April 13, 2025 at 4:24 AM
New preprint! We use generative diffusion models to quickly produce realistic #Euclid -like galaxies, a step towards addressing challenges in calibrating and validating weak lensing analyses. 🔭🧪
arxiv.org/abs/2504.07183
arxiv.org/abs/2504.07183
We got so busy I forgot to follow up, our team did amazing and we were recognized with the Rising All-Star Award! I'm very proud of everything we accomplished despite the challenges, from our first swerve year to the wildfires. #firstrobotics #frc
April 13, 2025 at 3:46 AM
We got so busy I forgot to follow up, our team did amazing and we were recognized with the Rising All-Star Award! I'm very proud of everything we accomplished despite the challenges, from our first swerve year to the wildfires. #firstrobotics #frc
So excited to get our first complete swerve, vision, and kinematics stack working!!! I'm very impressed by how easy it was to build with the wpilib and photonvision APIs. Yes, the camera is mounted upside down. Shoutout to 980 for letting us use their field on weekends 🤖 #FRC #robotics
March 3, 2025 at 6:59 PM
High-resolution airborne AVIRIS-3 L1b radiance is publicly available for the #LAwildfire via Earthdata, and visualized on MMGIS. Incredible imagery of burn extent. #remotesensing 🌎🛰️
popo.jpl.nasa.gov/mmgis-aviris...
www.earthdata.nasa.gov/data/alerts-...
popo.jpl.nasa.gov/mmgis-aviris...
www.earthdata.nasa.gov/data/alerts-...
January 16, 2025 at 7:51 PM
High-resolution airborne AVIRIS-3 L1b radiance is publicly available for the #LAwildfire via Earthdata, and visualized on MMGIS. Incredible imagery of burn extent. #remotesensing 🌎🛰️
popo.jpl.nasa.gov/mmgis-aviris...
www.earthdata.nasa.gov/data/alerts-...
popo.jpl.nasa.gov/mmgis-aviris...
www.earthdata.nasa.gov/data/alerts-...
Finally, we think we have the beginnings of an instrument agnostic model for imaging spectroscopy. Here's SpecTf trained only on #EMIT, an orbital instrument, detecting clouds on #AVIRIS-NG, an airborne instrument with a different spectral grid. Exciting!
January 10, 2025 at 4:40 AM
Finally, we think we have the beginnings of an instrument agnostic model for imaging spectroscopy. Here's SpecTf trained only on #EMIT, an orbital instrument, detecting clouds on #AVIRIS-NG, an airborne instrument with a different spectral grid. Exciting!
SpecTf's attention weights are interpretable - the model learned O2-A, H2O, and CO2 absorptions, all physically relevant for cloud detection. Look at how smooth these features are! Treating the spectra as a sequence instead of image channels or independent features is critical.
January 10, 2025 at 4:40 AM
SpecTf's attention weights are interpretable - the model learned O2-A, H2O, and CO2 absorptions, all physically relevant for cloud detection. Look at how smooth these features are! Treating the spectra as a sequence instead of image channels or independent features is critical.
With a single transformer encoder layer, we demonstrate a significantly improved cloud detection model that is *pixelwise*, purely learning from the spectra alone without spatial or temporal biases.
January 10, 2025 at 4:40 AM
With a single transformer encoder layer, we demonstrate a significantly improved cloud detection model that is *pixelwise*, purely learning from the spectra alone without spatial or temporal biases.
Our preprint on the Spectroscopic Transformer and all associated code and datasets are now available! SpecTf is a #deeplearning model specifically for imaging spectroscopy and #EMIT #remotesensing
Preprint: arxiv.org/abs/2501.04916
Code: github.com/emit-sds/Spe...
Dataset: doi.org/10.5281/zeno...
Preprint: arxiv.org/abs/2501.04916
Code: github.com/emit-sds/Spe...
Dataset: doi.org/10.5281/zeno...
January 10, 2025 at 4:40 AM
Our preprint on the Spectroscopic Transformer and all associated code and datasets are now available! SpecTf is a #deeplearning model specifically for imaging spectroscopy and #EMIT #remotesensing
Preprint: arxiv.org/abs/2501.04916
Code: github.com/emit-sds/Spe...
Dataset: doi.org/10.5281/zeno...
Preprint: arxiv.org/abs/2501.04916
Code: github.com/emit-sds/Spe...
Dataset: doi.org/10.5281/zeno...
I had an awesome Friday afternoon (!) poster session at #AGU24. Thanks to everyone that stayed until the end, and I'm looking forward to another year of exciting collaborations.
December 14, 2024 at 8:34 PM
I had an awesome Friday afternoon (!) poster session at #AGU24. Thanks to everyone that stayed until the end, and I'm looking forward to another year of exciting collaborations.
Looks cool in the dark 🌌
December 7, 2024 at 2:19 AM
Looks cool in the dark 🌌
I'm at #AGU24 next week! We're excited to share our new Spectroscopic Transformer for an improved #EMIT cloud mask (GC41H-0038, Thurs. poster) and preliminary #methane retrievals for anthropogenic plume detection (A53I-2209, Fri. poster). Come chat about deep learning for imaging spectroscopy!
December 6, 2024 at 9:46 PM
Could I get a:
There's no place like 127.0.0.1:8888/?token=js79j7...
There's no place like 127.0.0.1:8888/?token=js79j7...
December 1, 2024 at 12:28 AM
Could I get a:
There's no place like 127.0.0.1:8888/?token=js79j7...
There's no place like 127.0.0.1:8888/?token=js79j7...
The "AI Outline" feature built into the Google Scholar PDF Reader is actually pretty neat for going through papers, and follows a pattern I wish all other GenAI tools followed: out of the way until I want to use it.
November 22, 2024 at 6:04 AM
The "AI Outline" feature built into the Google Scholar PDF Reader is actually pretty neat for going through papers, and follows a pattern I wish all other GenAI tools followed: out of the way until I want to use it.
Our #FRC 2404 robotics students are VERY excited to be assembling their new Thrifty swerve drive modules! Hopefully the days of 6-wheel drive are behind us 🤖
November 15, 2024 at 2:23 AM
Our #FRC 2404 robotics students are VERY excited to be assembling their new Thrifty swerve drive modules! Hopefully the days of 6-wheel drive are behind us 🤖
Since Europa Clipper is up the hill at JPL for vibration testing, I no longer have an excuse to walk over to the high bay gallery to clear my head (aka procrastinate)... Instead, I'm sharing this pic I took last month!
November 8, 2023 at 2:53 AM
Since Europa Clipper is up the hill at JPL for vibration testing, I no longer have an excuse to walk over to the high bay gallery to clear my head (aka procrastinate)... Instead, I'm sharing this pic I took last month!